Fix typo in the tutorial

[iolib.git] / examples / tutorial.tmpl

             Network Programming in ANSI Common Lisp with IOLib
                   by: Peter Keller (psilord@cs.wisc.edu)
                               Version 0.0
                                4/02/2010


What is IOLib?
--------------

IOLib is a portable I/O library for ANSI Common Lisp. It includes socket
interfaces for network programming with IPV4/IPV6 TCP and UDP, an I/O
multiplexer that includes nonblocking I/O, a DNS resolver library, and a
pathname library.

Where do I get IOLib?
---------------------

The current version of IOLib is found here:

http://common-lisp.net/project/iolib/download.shtml

Please use the repository located in the Live Sources section for the most up
to date version of IOLib.

Introduction
------------

This tutorial loosely follows the exposition of network programming in "UNIX
Network Programming, Networking APIs: Sockets and XTI 2nd Edition" by W.
Richard Stevens. Many examples are derived from the source codes in that book.
Major deviations from the C sources include converting the concurrent examples
which use fork() into threaded examples which use the portable Bordeaux Threads
package, more structured implementations of certain concepts such as data
buffers and error handling, and general movement of coding style towards a
Common Lisp viewpoint.

The scope of this version of the tutorial is:
    0. Exposition suitable for programmers unfamiliar with ANSI Common Lisp
    1. IPV4 TCP
    2. Client/Server architecture
    3. Iterative vs Concurrent (via threading) vs Multiplexed Server Design
    4. Blocking and nonblocking I/O

It is intended, however, that this tutorial grows to contain the entirety of
IOLib's API as detailed in the Future Directions section of this tutorial. As
newer revisions of this tutorial are released, those gaps will be filled until
the whole of the IOLib API has been discussed.

Finally, the example code in this tutorial is algorithmically cut from the
actual example programs and inserted into the tutorial via a template
generation method. The example codes have embedded in them a tiny markup
language which facilitates this in the form (on a single line) of ';; ex-NNNb'
to begin an example section, and ';; ex-NNNe' to end an example section--NNN
stands for an enumeration integer for which each section's begin and end must
match.

Acknowledgements
----------------

I would like to greatly thank Stelian Ionescu, the author of IOLib
for his exposition of the various features of IOLib and his patience
in our sometimes long conversations.

Supporting Code
---------------

The file package.lisp contains a small library of codes used widely in the
examples. The supporting code implements:

    0. The package containing the examples, called :iolib.examples.

    1. The variables *host* and *port*, set to "localhost" and 9999
       respectively. This is the default name and port to which
       client connect and servers listen. Servers usually bind
       to 0.0.0.0, however.

    2. A small, but efficient, queue implementation, from "ANSI Common Lisp"
       by Paul Graham. The interface calls are:
       (make-queue)
       (enqueue obj q)
       (dequeue q)
       (empty-queue q)

    3. :iolib.examples currently depends upon IOLib alone and uses 
       packages :common-lisp, :iolib, and :bordeaux-threads.

Running the Examples
--------------------

These examples were developed and tested on SBCL 1.0.33.30 running on an x86
Ubuntu 8.10 machine. They were ran with two sessions of SBCL running, one
acting as a client, and the other as a server.

Supposing we'd like to start up the first example of the daytime server and
connect to it with the first daytime client example. Initially, the server will
bind to *host* and *port* and wait for the client to connect. We connect with
the client to *host* and *port*, get the time, and exit.

First we'll start up a server:

Linux black > sbcl
This is SBCL 1.0.33.30, an implementation of ANSI Common Lisp.
More information about SBCL is available at <http://www.sbcl.org/>.

SBCL is free software, provided as is, with absolutely no warranty.
It is mostly in the public domain; some portions are provided under
BSD-style licenses.  See the CREDITS and COPYING files in the
distribution for more information.
* (require :iolib.examples) ; much output!
* (in-package :iolib.examples)

#<PACKAGE "IOLIB.EXAMPLES">
* (run-ex1-server)
Created socket: #<passive IPv4 stream socket, unbound {BF84B99}>[fd=5]
Bound socket: #<passive IPv4 stream socket bound to 0.0.0.0/9999 {BF84B99}>
Listening on socket bound to: 0.0.0.0:9999
Waiting to accept a connection...
[ server is waiting for the below client to connect! ]
Got a connection from 127.0.0.1:34794!
Sending the time...Sent!
T
*

Now we'll start up the client which connected to the above server:

Linux black > sbcl
This is SBCL 1.0.33.30, an implementation of ANSI Common Lisp.
More information about SBCL is available at <http://www.sbcl.org/>.

SBCL is free software, provided as is, with absolutely no warranty.
It is mostly in the public domain; some portions are provided under
BSD-style licenses.  See the CREDITS and COPYING files in the
distribution for more information.
* (require :iolib.examples) ; much output!
* (in-package :iolib.examples)

#<PACKAGE "IOLIB.EXAMPLES">
* (run-ex1-client)
Connected to server 127.0.0.1:9999 via my local connection at 127.0.0.1:34794!
2/27/2010 13:51:48
T
*

In each client example, one can specify which host or port to which it should
connect:

* (run-ex1-client :host "localhost" :port 9999)
Connected to server 127.0.0.1:9999 via my local connection at 127.0.0.1:34798!
2/27/2010 13:53:7
T
*

The servers can be told a port they should listen upon and in this tutorial,
unless otherwise specified, will always bind to 0.0.0.0:9999 which means across
all interfaces on the machine and on port 9999.


                                   CHAPTER 1
                                   ---------
                             IPV4 TCP Client/Server
                          Blocking and nonblocking I/O

Overview of Examples
--------------------

The examples consist of a collection of clients and servers. They are split
into two groups: a set of daytime clients and server, and echo clients and
servers. In some of the examples, a certain network protocol, suppose
end-of-file handling, must be matched between client and server causing further
delineation.

Client protocols are matched to server protocols thusly:

Clients: ex1-client, ex2-client, ex3-client, can work
with servers: ex1-server, ex2-server, ex3-server, ex4-server.

Clients: ex4-client, ex5a-client, can work with servers: ex5-server,
ex6-server.

Clients: ex5b-client, can work with servers: ex7-server, ex8-server

Some clients and servers use the "daytime" series of protocols, those
are ex1-client, ex2-client, ex3-client, and ex1-server, ex2-server,
ex3-server, and ex4-server.

Some clients and servers use the "echo a line" series of protocols,
those are ex4-client, ex5a-client, ex5b-client, and ex5-server,
ex6-server, ex7-server, and ex8-server.

Even though much of the example source is included in the tutorial, it is
recommended that the example sources be carefully read and understood in order
to gain the most benefit from the tutorial.

Daytime Clients
---------------

In this section we show the evolution of a client which connects to a server
and gets the time of day. Each example shows some kind of an incremental
improvement to the previous one.

Daytime Client IVP4/TCP: ex1-client.lisp
----------------------------------------

This example is a very simple daytime client program which contacts a server,
by default at *host* and *port*, returns a single line of text that is the
current date and time, and then exits. It is written in more of a C style just
to make it easy to compare with similar simple examples in other languages. It
uses blocking, line oriented I/O.

The steps this program performs are:

0. The ex1-client.lisp entrance call:

    <example ex1-client:ex-0>

1. Create an active TCP socket:

        The socket creation function (MAKE-SOCKET ...) is the method by which one
        creates a socket in IOLib.  It is very versatile and can be used to both
        create and initialize the socket in a single call.

        In this case, we use it simply and create an active IPV4 Internet stream
        socket which can read or write utf8 text and that understands a particular
        newline convention in the underlying data.

        One small, but important, deviation of IOLib sockets from Berkeley sockets
        is that when a socket is created, it is predestined to forever and
        unalterably be either an active or passive socket. Active sockets are used
        to connect to a server and passive sockets are used for a server's
        listening socket.

    <example ex1-client:ex-1>

2. Specify the Server's IP address and port and establish a connection 
    with the server:

        This bit of code contains many calls into IOLib and we shall examine each
        of them. 

    The function LOOKUP-HOSTNAME takes as a string the DNS name
    for a machine and returns 4 values:
        A. an address
        B. a list of additional addresses(if existent)
        C. the canonical name of the host
        D. an alist of all the host's names with their respective addresses 

        We use only the first return value, the address component, to pass to the
        function CONNECT.

        The function CONNECT will connect the socket to the address, but to a
        random port if the :port keyword argument is not specified. The average
        client codes usually use :wait t to block until the connect can resolve
        with a connected fd or an error. The exception to always using :wait t is
        if the client needs to connect to many servers at once, suppose a web
        client, or if a server is also a client in other contexts and wishes not to
        block.

        The functions REMOTE-HOST and REMOTE-PORT return the ip address and port of
        the remote connection associated with the connected socket. LOCAL-HOST and
        LOCAL-PORT return the information of the client's end of the connected
        socket. Analogous calls REMOTE-NAME and LOCAL-NAME each return two values
        where the first value is the equivalent of *-host and the second value is
        the equivalent of *-port.

    <example ex1-client:ex-2>

3. Read and display the server's reply:

        Now that the socket has been connected to the server, the server will send
        a line of text to the client. The client uses the standard Common Lisp
        function READ-LINE to read the information from the socket. The function
        READ-LINE blocks and will only return when an *entire line* is read. Once
        read, the line is emitted to *standard-output* via the function call
        FORMAT.

    <example ex1-client:ex-3>

4. End program:

        We close the socket with the standard function CLOSE and return true so the
        return value of this example is t.

    <example ex1-client:ex-4>

While this program works, it has some major flaws in it. First and foremost is
that it doesn't handle any conditions that IOLib signals in common use cases.
An example would be to run the ex1-client.lisp example without a daytime server
running. In most, if not all, Common Lisp toplevels, you'll be dropped into the
debugger on an unhandled SOCKET-CONNECTION-REFUSED-ERROR condition. Secondly,
it isn't written in the Common Lisp style.

Daytime Client IVP4/TCP: ex2-client.lisp
----------------------------------------

In this example, we simply tackle the fact ex1-server.lisp can be shortened
with an IOLib form to something where the application writer has less to do
concerning cleaning up the socket object. It also uses line oriented blocking
I/O.

The introduced macro WITH-OPEN-SOCKET calls MAKE-SOCKET with the arguments in
question and binds the socket to the variable 'socket'. When this form returns,
it will automatically close the socket.

This shortens the program so much, that the example can be included in its 
entirety:

    <example ex2-client:ex-0>

This shorthand can go even further, if we add this to the WITH-OPEN-SOCKET
flags

    :remote-host (lookup-hostname host)
    :remote-port port

then the underlying MAKE-SOCKET call will in fact connect the socket directly
to the server before it is available for the body of the macro allowing us to
remove the connect call entirely! In the early examples, however, we don't
utilize IOLib's shorthand notations to this degree in order to make apparent
how the library maps into traditional socket concepts. After one gains
familiarity with the IOLib API, the situations where application of the
shortcuts are useful become much easier to see.

Daytime Client IVP4/TCP: ex3-client.lisp
----------------------------------------

Now we come to condition handling, which can moderately affect the layout of
your IOLib program. Any real program using IOLib must handle IOLib's signaled
conditions which are common to the boundary cases of network programming.
We've already seen one of these boundary cases when we tried to connect a
daytime client to a server that wasn't running.  The condition signaled in that
case was: SOCKET-CONNECTION-REFUSED-ERROR.  The stream interface has a set of
conditions which IOLib will signal, and another lower level IOLib layer--which
we'll come to in the nonblocking I/O examples have another set of conditions.
There is some intersection between them and we will explore that later. For
now, we'll just use the conditions associated with a stream.

Our rewrite of ex2-client.lisp into ex3-client.lisp (continuing to use line
oriented blocking I/O) proceeds thusly:

0. We create a helper function which connects to the server and reads the
    daytime line:

        Notice the HANDLER-CASE macro around the portion of the function which
        reads the date from the server. In looking at the boundary conditions from
        the server given this protocol, we can receive an END-OF-FILE condition if
        the client connected, but before the server could respond it exited,
        closing the connection. Since in this case we're inside of a
        WITH-OPEN-SOCKET form, we simply note that we got an END-OF-FILE and let
        the cleanup forms of WITH-OPEN-SOCKET close the connection. If we don't
        catch this condition, then the program will break into the debugger and
        that isn't useful.  It is usually debatable as to where one should handle
        conditions: either near to or far away from the generating calls. In these
        simple examples, no choice has any significant pros or cons. As your IOLib
        programs become more and more complex, however, it becomes more obvious at
        what abstraction level to handle signaled conditions.

    <example ex3-client:ex-0>

1. Some conditions which are complete show-stoppers to the functioning of the
    code are caught at a higher level:

        Notice we catch the possible SOCKET-CONNECTION-REFUSED-ERROR from the
        connect inside of the function run-ex3-client-helper.

    <example ex3-client:ex-1>

Here are some common conditions in IOLib (some from ANSI Common Lisp too) and
under what situations they are signaled.  In any IOLib program, *at least*
these conditions should be handled where appropriate.

END-OF-FILE:
    When a stream function such as READ, READ-LINE, etc...(but not
    RECEIVE-FROM), reads from a socket where the other end has been closed.

HANGUP:
    When writing to a socket with a stream function such as WRITE,
    FORMAT, etc...(but not SEND-TO), if the socket is closed then this
    condition is signaled.

SOCKET-CONNECTION-RESET-ERROR:
    When doing I/O on a socket and the other side of the socket sent a
    RST packet, this condition is signaled.  It can also happen with
    the IOLIb function ACCEPT and similar.

SOCKET-CONNECTION-REFUSED-ERROR:
    Signaled by connect if there is no server waiting to accept the incoming
    connection.


Daytime Servers
---------------

Now that we have completed the evolution of the daytime client, let's look at
the daytime servers.

The exposition of the servers follows in style of the clients.

Daytime Server IVP4/TCP: ex1-server.lisp
----------------------------------------

This first example is an iterative server which handles a single client and
then exits. The I/O is blocking and no error handling is performed.  This is
similar in scope to the ex1-client.lisp example.

0. Create the server socket:

        We see that the socket is :passive. Every socket in IOLib is predestined to
        be either an active or passive socket and since this is a server socket, it
        is passive. Also here we see that we can ask for the underlying fd of the
        socket with the function SOCKET-OS-FD.

    <example ex1-server:ex-0>

1. Bind the socket

        Binding a socket is what gives it an endpoint to which clients can connect.
        The IOLib constant +IPV4-UNSPECIFIED+ represents 0.0.0.0 and means if a
        connection arrives on any interface, it will be accepted if it comes to the
        :port specified. The :reuse-addr keyword represents the socket option
        SO_REUSEADDR and states (among other things) that if the socket is in the
        TIME_WAIT state it can be reused immediately.  It is recommended that all
        servers use :reuse-addr on their listening socket.

    <example ex1-server:ex-1>

2. Listen on the socket

        Listening on a socket allows clients to connect. In this example, we've
        specified that 5 pending connection can be queued up in the kernel before
        being accepted by the process.

    <example ex1-server:ex-2>

3. Accept the client connection.

        Here we finally call the IOLib function ACCEPT-CONNECTION. We would like it
        to block, so we pass it :wait t. When ACCEPT-CONNECTION returns it will
        return a new socket which represents the connection to the client.
        ACCEPT-CONNECTION can return nil under some situations, such as on a slow
        server when the client sent a TCP RST packet in between the time the kernel
        sees the connection attempt and ACCEPT-CONNECTION is actually called.  We
        also opt to use the function REMOTE-NAME, which returns two values, the ip
        address and port of the remote side of the socket.

    <example ex1-server:ex-3>

4. Write the time to the client.

        Here we've figured out the time string and wrote it to the client.  Notice
        we call the function FINISH-OUTPUT. This ensures that all output is written
        to the client socket. For streams using blocking I/O, it is recommended
        that every write to a blocking socket be followed up with a call to
        FINISH-OUTPUT.

    <example ex1-server:ex-4>

5. Close the connection to the client.

    We're done writing to the client, so close the connection so the client
        knows it got everything.

    <example ex1-server:ex-5>

6. Close the server's socket.

        Since this is a one shot server, we close the listening socket and exit. In
        this and all other servers we call FINISH-OUTPUT to flush all pending
        message to *standard-output*, if any.

    <example ex1-server:ex-6>

The above code is the basic idea for how a very simple TCP blocking I/O server
functions. Like ex1-client, this server suffers from the inability to handle
common signaled conditions such as a HANGUP from the client--which means the
client went away before the server could write the time to it.

However, one major, and subtle, problem of this particular example is that the
socket to the client is *not immediately closed* if the server happens to exit,
say by going through the debugger back to toplevel--or a signaled condition,
before writing the date to the client. If this happens, it can take a VERY long
time for the socket to be garbage collected and closed. In this scenario, the
client will hang around waiting for data which will never come until the Lisp
implementation closes the socket when it gets around to collecting it. Garbage
collection is an extremely nice feature of Common Lisp, but non-memory OS
resources in general should be eagerly cleaned up.  Clients can suffer from
this problem too, leaving open, but unmanipulable, sockets to servers.

All clients or servers written against IOLib should either use some IOLib
specific macros to handle closing of socket, Common Lisp's condition system
like handler-case to catch the signaled conditions, or some other manual
solution.

Daytime Server IVP4/TCP: ex2-server.lisp
----------------------------------------

Similarly to ex2-client, this server uses the macro WITH-OPEN-SOCKET to open
the server socket. We introduce WITH-ACCEPT-CONNECTION to accept the client and
convert this server from a single shot server to an iterative server which can
handle, in a serial fashion only, multiple clients.

0. Serially accept and process clients:

        This portion of ex2-server shows the infinite loop around the accepting of
        the connection.  The macro WITH-ACCEPT-CONNECTION takes the server socket
        and introduces a new binding: client, which is the accepted connection. We
        ensure to tell the accept we'd like to be blocking. If for whatever reason
        we exit the body, it'll clean up the client socket automatically.

    <example ex2-server:ex-0>

For very simple blocking I/O servers like this one, serially accepting and
handling client connections isn't so much of a problem, but if the server does
anything which takes a lot of time or has to send lots of data back and forth
to many persistent clients, then this is a poor design. The means by which you
exit this server is by breaking evaluation and returning to the toplevel. When
this happens, the WITH-* forms automatically close the connection to the
client.

Daytime Server IVP4/TCP: ex3-server.lisp
----------------------------------------

In this iterative and blocking I/O server example, we add the handling of the
usual signaled conditions in network boundary cases often found with sockets.
Like the earlier client where we introduced HANDLER-CASE, this involves a
little bit of restructuring of the codes.

0. A helper function which opens a passive socket, binds it, and
    listens on it:

        There is nothing new in this portion of the code. We've seen this pattern
        before. In production code, we could probably shorten this further by
        having WITH-OPEN-SOCKET do the binding and connecting with appropriate
        keyword arguments.

    <example ex3-server:ex-0>

1. Repeatedly handle clients in a serial fashion:

        The new material in this function is the HANDLER-CASE around sending the
        client the time information. The boundary conditions when writing to a
        client include the server getting a reset (RST) from the client or
        discovering the client had gone away and there is no-one to which to write.
        Since the write is contained within the WITH-ACCEPT-CONNECTION form, if any
        of these conditions happen, we simply notice that they happened and let the
        form clean up the socket when it exits.  If we didn't catch the conditions,
        however, we'd break into the debugger.

        One might ask what the value of catching these conditions here is at all
        since we don't actually do anything with them--other than printing a
        message and preventing the code from breaking into the debugger. For the
        purposes of the tutorial, it is intended that the reader induce the
        boundary cases manually and see the flow of the code and to understand
        exactly what conditions may be signaled under what conditions and how to
        structure code to deal with them. In production code where the author might
        not care about these conditions at all, one might simply ignore all the
        signaled conditions that writing to the client might cause.

        Of course, the appropriateness of ignoring network boundary conditions is
        best determined by context.

    <example ex3-server:ex-1>

2. End of the helper function, returns T to whomever called it:

    <example ex3-server:ex-2>

3. The entry point into this example:

        We handle the condition SOCKET-ADDRESS-IN-USE-ERROR which is most commonly
        signaled when we try to bind a socket to address which already has a server
        running on it or when the address is in the TIME_WAIT state. The latter
        situation is so common--usually caused by a server just having exited and
        another one starting up to replace it, that when binding addresses, one
        should supply the keyword argument :reuse-addr with a true value to
        BIND-ADDRESS to allow binding a socket to an address in TIME_WAIT state.

    <example ex3-server:ex-3>

Daytime Server IVP4/TCP: ex4-server.lisp
----------------------------------------

This is the first of our concurrent servers and the last of our daytime
protocol servers. Usually concurrency is introduced (in the UNIX environment)
with the fork() library call which creates an entirely new process with
copy-on-write semantics to handle the connection to the client. In this
tutorial environment, we've chosen to render this idea with the portable
threading library Bordeaux Threads.  The I/O is still line oriented and
blocking, however, when a thread blocks another can run giving the illusion of
a server handling multiple clients in a non-blocking fashion.

We also introduce UNWIND-PROTECT ensures that various sockets are closed under
various boundary conditions in the execution of the server.  An UNWIND-PROTECT
executes a single form, and after the evaluation, or interruption, of that
form, evaluates a special cleanup form. The cleanup form is *always* evaluated
and we use this to cleanup non-memory system resources like sockets.

Threads present their own special problems in the design of a server. Two
important problems are: data races and thread termination. The tutorial tries
very hard to avoid any data races in the examples and this problem is
ultimately solvable using Bordeaux-Threads mutexes or condition variables.  Our
simple examples do not need mutexes as they do not share any data between
themselves. 

The harder problem is thread termination. Since the tutorial encourages
experimentation with the clients and servers in a REPL, threads may leak when
the server process' initial thread stops execution and goes back to the REPL.
We use three API calls from the Bordeaux Threads: THREAD-ALIVE-P, ALL-THREADS,
and DESTROY-THREAD--which are not to be used in normal thread programming.  We
do this here in order to try and clean up leaked threads so the clients know
immediately when the server process stopped and we don't pollute the REPL with
an ever increasing number of executing threads. The employed method of
destroying the threads, on SBCL specifically, allows the invocation of the
thread's UNWIND-PROTECT's cleanup form, which closes the socket to the client
before destroying the thread.  On other implementations of Common Lisp, we are
not guaranteed that the thread's UNWIND-PROTECT cleanup form will be evaluated
when we destroy it.

This method is also extremely heavy handed in that it uses the function
IGNORE-ERRORS to ignore any condition that Bordeaux Thread's DESTROY-THREAD may
have signaled, including important conditions like HEAP-EXHAUSTED-ERROR, an
SBCL specific condition. In a real threaded server, the exiting of the initial
thread (which means exiting of the runtime and termination of the entire Lisp
process) will destroy all other threads as the process tears itself down and
exits. This is the recommended way a threaded server should exit.

Since threading is implementation dependent for what guarantees are provided,
any non-toy threaded network server will probably use the native implementation
of threads for a specific Common Lisp implementation.  An example difficult
situation would be trying to terminate a thread which is blocked on I/O.
Different implementations would handle this in different ways.

The two provided examples, ex4-server and ex5-server, provide a general idea
for the structuring of the code to utilize threads.

Here is the dissection of ex4-server:

0. A special variable which will allow the initial thread to pass a client
    socket to a thread handling said client:

    <example ex4-server:ex-0>

1. A helper function which begins with the usual recipe for a server:

    <example ex4-server:ex-1>

2. Forever more, accept a client connection on the listening socket
    and start a thread which handles it:

        There is a lot going on in this piece of code. The first thing to notice is
        the UNWIND-PROTECT and its cleanup form. The form which UNWIND-PROTECT is
        guarding is an infinite loop which does a blocking accept to get a client
        socket, rebinds *default-special-bindings* adding to its assoc list the
        binding for *ex4-tls-client*, and creates a thread which handles the
        client.

        The cleanup form walks all of the active client threads and destroys them,
        ignoring any conditions that may have arose while doing so. Destroying the
        threads prevents them from piling up and eventually causing havoc if many
        servers start and exit over time. In addition, it forces an eager close on
        the client sockets allowing any clients to know the server went away
        immediately.

    <example ex4-server:ex-2>

3. The beginning of the thread handling the client:

        When the thread is born, the aforementioned explicit binding of the client
        socket to *ex4-tls-client* takes effect via the *default-special-bindings*
        mechanism. By declaring *ex4-tls-client* ignorable, we inform the compiler
        that this variable is set "elsewhere" and no warning should be emitted
        about its possibly undefined value. In our case, this will always be
        defined at runtime in this server.

    <example ex4-server:ex-3>

4. Send the time to the socket:

        The UNWIND-PROTECT in this form handles every possible case of leaving the
        evaluable function such as it completing normally, a condition being
        signaled, or by thread destruction--on SBCL! In all cases, the socket to
        the client is closed which cleans up OS resources and lets the client know
        right away the server has closed the connection. The HANDLER-CASE form here
        just informs us which of the common IOLib conditions may have been signaled
        while writing the time to the client.

    <example ex4-server:ex-4>

        It is a bit tricky to robustly handle closing of the client socket in the
        thread. For example, if we bound the special variable *ex4-tls-client* to a
        lexically scoped variable and then did the UNWIND-PROTECT form to close the
        lexically scoped variable, then if this thread wakes up and gets destroyed
        after the lexical binding, but before the UNWIND-PROTECT, we'd lose a
        socket to a client into the garbage collector.

    Such incorrect code would look like:

    ;; This code is incorrect!
    (defun process-ex4-client-thread ()
      (declare (ignorable *ex4-tls-client*))
      (let ((client *ex4-tls-thread*))
        ;; thread gets destroyed right here! client socket is left open!
        (unwind-protect
          ( <evaluable form> )
          (close client))))

5. The entry point into this example:

        Like earlier servers, we call the helper function and catch what happens if
        :reuse-addr wasn't true in the BIND-ADDRESS function call.

    <example ex4-server:ex-5>


Daytime Client/Server Commentary
--------------------------------

This concludes the examples using the daytime protocol. We've seen patterns
emerge in how the simplest of clients and servers are built and began to reason
about how to handle common signaled conditions. Threading, of course, increases
the care one must have in order to ensure that data access and control flow is
kept consistent.

Echo Line Clients and Servers
-----------------------------

These next examples focus on the echo protocol. This is simply a server that
sends back to the client whatever the client wrote to it.  A client can request
to quit talking to a server (except ex7-server and ex8-server, where this
feature isn't implemented) by sending the word "quit", on a line by itself.
This tells the server to close the connection to the client once it has
finished echoing the line. The closing of the client's read socket lets the
client know the connection to the server went away and that it is time to exit.
We also introduce the socket multiplexer interface which allows concurrent
processing of socket connections. This is similar to how UNIX's select(),
epoll(), or kqueue() works. Due to portability concerns on doing nonblocking
operations on *standard-input* and *standard-output* (we can't easily do it) we
are beholden to some form of blocking I/O in our clients because they interact
with a human. We will explore true non-blocking I/O in the ex8-server example
since that server only has to converse with connected clients.

Echo Clients
------------

The echo clients are a group of programs which read a line from
*standard-input*, write it to the server, read back the response from the
server, and emit the result to *standard-output*.  While there is a portable
method to read "however much is available" from *standard-input*, there isn't
the symmetrical method to write "whatever I'm able" to *standard-output*.  For
our client design, this means that all of these clients are line oriented and
do blocking I/O when reading from *standard-input* and writing to
*standard-output*.

Echo Client IPV4/TCP: ex4-client.lisp
--------------------------------------

This is a very basic echo client program that handles the usual conditions
while talking to the server:

0. Connect to the server and start echoing lines:

        Here we use WITH-OPEN-SOCKET to create an active socket that we then use to
        connect to the server. We handle HANGUP, for when the server went away
        before the client could write to it, and END-OF-FILE, for when the server
        closes down the connection.

        Notice we call the function ex4-str-cli inside of a HANDLER-CASE macro.
        This allows us to not check for any signaled conditions in ex4-str-cli and
        greatly simplifies its implementation.

        In this specific example, we don't do anything other than notify that the
        condition happened since after that the socket gets closed via
        WITH-OPEN-SOCKET.

    <example ex4-client:ex-0>

1. Echo lines to the server:

        Until the user inputs "quit" on a line by itself, we read a line, send it
        to the server, read it back, and emit it to stdout. If any of the usual
        conditions are signaled here, the handler-case in the Step 0 code fires and
        we deal with it there.

        When "quit" is entered, the line is sent on the round trip to the server
        like usual, but this time the server closes the connection to the client.
        Unfortunately, since the client is doing blocking I/O, we must read another
        line from *standard-input* before we get any signaled condition when IOLib
        discovers the socket has been closed by the server.

        In practice, this means after the server closed the connection, the user
        must hit <return> in order to drive the I/O loop enough to get the signaled
        condition.

    <example ex4-client:ex-1>

2. Entry point into the example:

        We handle the usual connection refused condition, but otherwise this step
        is unremarkable.

    <example ex4-client:ex-2>

Echo Client IPV4/TCP: ex5a-client.lisp
--------------------------------------

This is the first client to use the socket multiplexer to notice when the
socket to the server is ready for reading or writing. While the multiplexer is
often used in single threaded servers it can be used for clients--especially
clients which may talk to multiple servers like web clients.  Use of the
multiplexer API will require a significant change in how the code is
structured. It is not recommended that the multiplexer and threads be used
simultaneously to handle network connections.

Keeping in mind the fact that we ALWAYS could block while reading from
*standard-input* or writing to *standard-output*, we only attempt to read/write
to the standard streams when the multiplexer thinks it can read/write to the
server without blocking. This is a change from the traditional examples of how
to do this in C because in C one can determine if STDIN or STDOUT are ready in
the same manner as a network file descriptor.

The first big change from our previous examples is that we stop using
WITH-OPEN-SOCKET since now we must manually control when the socket to the
server must be closed. This is especially important for clients who use active
sockets. The second change is how we do the creation and registering of the
handlers for reading and writing to the server socket.  The third change is how
to unregister a handler and close the socket associated with it under the right
conditions. Other changes will be explained as we meet them.

The main functions of the multiplexer API are:
    (make-instance 'iomux:event-base ....)
        Create an instance of the event-base, and associate some properties
        with it, such as event-dispatch should return if the multiplexer
        does not have any sockets it is managing.
        Passed an:
            :exit-when-empty - when no handlers are registered, event-dispatch
                                will return.

    (event-dispatch ...)
        By default, sit in the multiplexer loop forever and handle I/O
        requests. It is passed the event-base binding and in addition:
            :once-only - run the ready handlers once then return.
            :timeout - when there is no I/O for a certain amount of time return.

    (set-io-handler ...)
        Associates a handler with a state to be called with a specific socket.
        Passed an:
            event-base binding
            :read or :write or :error keyword
            the handler closure

    (remove-fd-handlers ...)
        Removes a handler for a specific state with a specific socket.
        Passed an:
            event-base binding
            an fd
            one or more of :read t, :write t, :error t

Here is the example using this API.

0. The event base:

        The event-base is the object which holds the state of the multiplexer.  It
        must be initialized and torn down as we'll see in the entry function to
        this example.

    <example ex5a-client:ex-0>

1. A helper function in which we create the active socket:

    Instead of using WITH-OPEN-SOCKET, we manually create the socket. We do
    this to better control how to close the socket. WITH-OPEN-SOCKET will try
    to FINISH-OUTPUT on the socket before closing it. This is bad if the socket
    had been previously closed or signaled a condition like HANGUP. Trying to
    write more data to an already hung up socket will simply signal another
    condition. To prevent layers of condition handling code, we explicitly
    handle closing of the socket ourselves.

    <example ex5a-client:ex-1>

2. Connect to the server, register the socket handlers:

        We protect the closing of the socket via UNWIND-PROTECT. We will talk about
        the ramifications of this decision in the next step which describes the
        UNWIND-PROTECT's cleanup form. In this section of code, we set up a read
        and write handler for the socket, and invoke the dispatch function, which
        will continue calling the handlers associated with the socket until the
        socket gets closed and the handlers unregistered. When this happens (see
        the entrance function step for why), EVENT-DISPATCH returns and we continue
        on to the cleanup form for the UNWIND-PROTECT.

    Setting up a handler in the multiplexer requires several arguments to
    the function set-io-handler. Here are what the arguments to that function
    are:
        1. *ex5a-event-base*
            This is the instance of the multiplexer for which we are setting
            up the handler.
        2. (socket-os-fd socket)
            This call returns the underlying operating system's file
            descriptor associated with the socket.
        3. :read
            This keyword states that we'd like to call the handler when the
            socket is ready to read. There is also :write and :error.
        4. (make-ex5a-str-cli-read    socket
                                    (make-ex5a-client-disconnector socket))
            The make-ex5a-str-cli-read function returns a closure over the
            socket and another closure returned by the
            make-ex5a-client-disconnector function. This function is what will
            be called when the socket is ready for reading. We will shortly
            explain the signature of this function and what gets passed to it
            by the multiplexer. The disconnector function will be called by the
            returned reader function if the reader function thinks that it
            needs to close the socket to the server.

    <example ex5a-client:ex-2>

3. Cleanup form for UNWIND-PROTECT:

        In the cleanup form, we always close the socket and we pass the function
        close :abort t to try and close the socket in any way possible. If we just
        tried closing the socket, then we might cause another condition to be
        signaled if a previous condition, like HANGUP, had already affected the
        socket. :abort t avoids that case. If the socket is already closed by a
        handler by the time we get here, closing it again hurts nothing.

    <example ex5a-client:ex-3>

4. Make the writer function for when the socket is ready to write:

        This function returns a closure which is called by the multiplexer when it
        is ready to read something from the server. The arguments to the closure
        are fd, the underlying file descriptor for the ready socket, event, which
        could be :read, :write, or :error if the handler was registered multiple
        times, and exception, which is nil under normal conditions, :error under an
        error with the socket, or :timeout, if we were using timeout operations
        when dealing with the socket.

        The closure will read a line with the function READ-LINE and write it to
        the server. The read will be blocking, but hopefully the write won't be
        since the multiplexer told us we could perform the write and not block.
        Obviously, is we write an enormous line, then we might block again, and in
        this case the FINISH-OUTPUT on the socket will push the data in a blocking
        I/O fashion until it is done and we return from the handler. So while this
        closure for the most part writes when ready, there are cases under which
        it'll still block.

        In this handler, if there is a signaled condition either reading from
        *standard-input* (the END-OF-FILE condition) or writing to the server
        socket (the HANGUP condition), we invoke the disconnector closure and pass
        it :close. When we get to the description of the disconnector function,
        you'll see what that means.

        Once the disconnector closure is invoked, the handler will have been
        removed and the socket closed. This will make EVENT-DISPATCH return since
        the only socket it was multiplexing for was closed--because we've told the
        multiplexer to do so when it was made!

    <example ex5a-client:ex-4>

5. Make the reader function for when the socket is ready to read:

        This piece of code is very similar to the previous step's code, we just
        handle the appropriate conditions and after reading the line from the
        server emit it to *standard-output*. Again, even though we are told we can
        read from the server without blocking, if the read is large enough we will
        continue to block until read-line reads the all the data and the newline.

    <example ex5a-client:ex-5>

6. The disconnector function:

        This function returns a closure which takes an arbitrary number of
        arguments. If the arguments to the invoked closure contain :read, :write,
        or :error, the respective handler on the associated socket is removed. If
        none of those three are supplied, then all handlers for that socket are
        removed.  Additionally if :close is specified, the socket is closed.  While
        not all features of this function is used in this example, this function
        (or a similar one using the correct event-base special variable) is used
        whenever we use the multiplexer in an example.

        The closure is called whenever a handler believes it should unregister
        itself or another handler, or close the socket. Because we will often close
        the socket in the disconnector closure, we can't use WITH-OPEN-SOCKET to
        automatically close the socket because WITH-OPEN-SOCKET may try to flush
        data on the socket, signaling another condition.

    <example ex5a-client:ex-6>

7. The entry point for this example and setting up the event-base:

        This function is much more complex than in examples that do not use the
        multiplexer. Protected by an UNWIND-PROTECT, we first initialize the event
        base my calling make-instance 'iomux:event-base.  Here is where we pass the
        keyword argument :exit-when-empty t which states that the event-dispatch
        function should return when there are no more registered handlers. Once
        that is done, we call the helper, catching a common condition and waiting
        until we return.

    <example ex5a-client:ex-7>

8. The cleanup form for UNWIND-PROTECT:

        This cleanup form closes the *ex5a-event-base* instance. IOLib defines a
        method for the generic function CLOSE which accepts an event-base and
        performs the necessary work to shut it down.

    <example ex5a-client:ex-8>

While this program works just fine with human input, it has a failure when
reading batch input. The failure is that when we get the END-OF-FILE condition
when *standard-input* closes, we _immediately_ unregister the read/write
handlers to the server, close the socket and exit the program. This destroys
any in-flight data to/from the server and lines being echoed may be lost.


Echo Client IPV4/TCP: ex5b-client.lisp
--------------------------------------

In order to fix the batch input problem of ex5a-client, we will use the
shutdown function which allows us to inform the server we are done writing
data, but leave the socket open so we can read the rest of the responses from
the server. This effectively closes only one-half of the TCP connection. The
server has to be made aware of this kind of protocol so it doesn't assume the
client completely exited when it gets an END-OF-FILE from the client and shuts
down the whole connection throwing away any queued data for the client.

This client is nearly identical to ex5a-client except we shut down the write
end of the connection to the server when we get END-OF-FILE from
*standard-input* and wait until we get all of the data back from the server.
The server signifies to us that it has sent all of the pending data by closing
the write end of its connection. The client sees the closing of the server's
write end as an END-OF-FILE on the socket connected to the server.

We show this example as a difference to ex5a-client.

0. Shutdown the write end of the socket to the server:

        Here we use the expanded functionality of the disconnector closure.  After
        we shut down the write end of our TCP connection, we call (funcall
        disconnector :write) which states only to remove the write (to the server)
        handler, but leave the connection open. After this happens, there is no way
        we can read from *standard-input* again.  Once the server sends the final
        data and the closes its connection to this client, we remove the read
        handler, which removes the last handler, and causes the EVENT-DISPATCH
        function to return, which ends the client computation.

    <example ex5b-client:ex-0>

Be aware that even if both directions on one end of a connection are shutdown,
close still must be called upon the socket in order to release resources held
by the operating system.

Echo Servers
------------

The echo servers, paired to clients as per the beginning of this tutorial,
further evolve to using the multiplexer and becoming more fine grained with
respect to when I/O is done until we reach the ability to perform nonblocking
I/O of arbitrary read/write sizes.

Echo Server IPV4/TCP: ex5-server.lisp
-------------------------------------

This threaded server is very similar to ex4-server, but instead of sending only
the time, each thread handles an echo protocol to a client.  While this is
still a blocking I/O server, only a single thread talking to a client gets
blocked, not the whole server.  Other than the server not honoring batch input
from the client correctly, this is a common model for a class of servers due to
its nonblocking behavior.

0. The special variable used to communicate the client socket to the thread:

    <example ex5-server:ex-0>

1. The usual prologue to a server:

    <example ex5-server:ex-1>

2. First half of creating the client threads:

    <example ex5-server:ex-2>

3. Second half, the cleanup form for the UNWIND-PROTECT:

    We make sure to clean up only the client threads!

    <example ex5-server:ex-3>

4. Handle the client and deal with signaled conditions:

        In this function, we ensure that under all conditions of the execution of
        this function, if something goes wrong, we eagerly close the socket to the
        client so it is not leaked into the garbage collector.  We also handle
        numerous conditions the the client could generate while talking to it in
        the function str-ex5-echo.

    <example ex5-server:ex-4>

5. Actually perform the echo protocol to the client:

        Read lines from the client and echo them back. All of this I/O is blocking.
        If we see "quit" from the client, then exit the loop, which causes the
        UNWIND-PROTECT cleanup form in step 4 to fire and close the connection to
        the client.

    <example ex5-server:ex-5>

6. The entrance function into this example:

    <example ex5-server:ex-6>


Echo Server IPV4/TCP: ex6-server.lisp
-------------------------------------

This is the first of the echo servers which use the multiplexer to handle
multiple clients concurrently. It is a single threaded program. As mentioned
before, one shouldn't mix the multiplexer and threads together to handle
network connections.

We explore a new concept with the multiplexer in that the listening server
socket is itself registered with the multiplexer. The read handler (called the
listener handler in this context) associated with this socket becomes ready
when a client has connected to the server address. Thus, once the listening
socket is ready the listener handler accepts the client and associates the line
echo protocol callback with the client's socket in the multiplexer.

The I/O design of this server is such that if the client connection is ready to
read, we read a line, then immediately write the line back to the client in the
same function without waiting to see if it is ready for writing. Since we are
still using blocking I/O, this is ok.  The reason for this example's design was
to minimize the complexity of using the multiplexer in order to introduce the
listener handler. Later examples become much more complex as we push the
multiplexer API farther.

0. The variable which holds the multiplexer instance:

    <example ex6-server:ex-0>

1. A hash table of client connections:

        We record each client that connects to the server into a hash table socket
        keyed by the list (ip address port) and associate with it a value of the
        client's socket. This is so that under any conditions of the server exiting
        we can eagerly close any open connections to clients in a cleanup form.

    <example ex6-server:ex-1>

2. Create and bind the server socket:

        We protect how we manipulate the server socket with an UNWIND-PROTECT so we
        ensure to close the socket at the end of the server's computation or if
        something went wrong.

    <example ex6-server:ex-2>

3. Register a listener handler on the server socket and start dispatching
    events with the multiplexer:

    <example ex6-server:ex-3>

4. When the server stops handling clients, we close the server socket:

    <example ex6-server:ex-4>

5. The listener handler:

        Once the returned closure from this function is called by the multiplexer
        on the ready server socket, we accept the client with a blocking accept.
        We then save the client connection in our table and register the line echo
        closure with the socket.  The line echo closure will also contain a
        disconnector function as in previous usages of the multiplexer.

    <example ex6-server:ex-5>

6. The line echo closure generator:

        This function returns a closure which is then bound to a client socket in
        the multiplexer. When the socket is ready, we read a line form the client
        and write it back to the client immediately. Since this is blocking I/O the
        whole server will wait until this transaction is complete.  This means that
        a client which sends one byte of ASCII that is not a newline can cause the
        whole server to block for all clients. This serious defect is remedied with
        non-blocking I/O, which we show in a later example.

    <example ex6-server:ex-6>

7. The disconnector closure generator:

        This function returns a closure that removes all the handlers from the
        socket in question and then closes it. Notice that this means this server
        is not capable of handling batch input from a client, since when it
        receives the END-OF-FILE on the read from a client, will immediately tear
        down the connection destroying any in flight data. After closing the
        socket, we also remove it from our table of open connections.

    <example ex6-server:ex-7>

8. Initialize the event-base, the connection table, and start the server:

        This code is the beginning of the UNWIND-PROTECT form which protects the
        server's socket resources.

    <example ex6-server:ex-8>

9. Cleanup the client connections and close the event-base:

        When the server exits we walk the *ex6-server-open-connections* hash and
        eagerly close every client we find there. After we are done, we close the
        event-base. This ensures every thing is cleaned up properly.

    <example ex6-server:ex-9>

This server uses the multiplexer in a simple fashion because only one handler
is registered for a client. That handler reads, then writes the data back to
the client.  The scope of the data read from the client never has to leave the
handler function.

Echo Server IPV4/TCP: ex7-server.lisp
-------------------------------------

This example is different than ex6-server because it fully separates the
reading and writing of data to a client into different handler functions. This
requires an architectural change to the server in order to be able to keep the
data from the client "somewhere" before being able to write it back to the
client when the multiplexer determines it can written to the client. We
introduce an io-buffer object, implemented in terms of a closure and one per
client, which stores the in-flight data until the client is ready to accept the
writes from the server.

Storage of client data introduces a problem in that if the client writes lots
of data to the server but happens to never be ready to accept it back from the
server, the server will consume all memory and run out of resources.  We
attempt to prevent this from happening, though not perfectly.

When the io-buffer is created for a client, we state we only would like a
certain number of bytes to be read from the client. Of course, since we're
using read-line with blocking I/O and the client could write a tremendous
amount of data before a newline, we can't completely enforce our storage policy
in this server. If the client, though, is well-behaved in that it sends
reasonable sized lines of text--a rarity in the real world, our implemented
policy is sufficient. When we reach the nonblocking I/O server example, we'll
find that we can perfectly enforce the per client data storage policy.

This server honors batch input from the client. When it sees the END-OF-FILE
from the client, and it still has data to write, the server will attempt to
write the rest of the data out as the multiplexer says the client is ready to
receive it.

Since this example is quite long the server portion will just be shown as a
difference to ex6-server.

0. The listener handler:

        The important code in this function is the call to make-ex7-io-buffer.
        This function returns a closure, here called io-buffer, which takes one
        argument, either :read-a-line or :write-a-line. When the funcall of
        io-buffer with the appropriate argument happens, *another* closure is
        returned and this is the closure registered with the appropriate ready
        state in the multiplexer.

        This returned closure has bound in its lexical scope the storage needed for
        the client.

        Both closures returned by :read-a-line and :write-a-line have access to the
        same storage space unique to this object io-buffer. This is the means by
        which the client's write handler can get access to the data read by the
        client's read handler.

    <example ex7-server:ex-0>

1. The disconnector function:

        This function is almost identical to a previous example used in
        ex5a-client.  The only difference is the special variable it references.

        Since the io-buffer knows under what conditions it should register or
        unregister specific handlers for the client socket, we need to be able to
        selectively remove them without disturbing the others.

    <example ex7-server:ex-1>

Now we come to the description of the ex7-io-buffer code base. This code base
interacts directly with the event-base multiplexer instance in order to
register and unregister handlers to the client. Handlers are only registered
when there is data to write, or room to read more data up to the buffer size.

0. The io-buffer closure generator and associated lexical storage:

        These are the variables closed over which represent the internal state of
        the closure and hold the data from the client. In particular note is the
        fact we keep track of when a handler is registered (since this object can
        register or unregister the handlers in and of itself) and whether or not
        we've seen the END-OF-FILE from a client. The line-queue will hold the
        actual data from the client.

    <example ex7-buffer:ex-0>

1. The read-a-line closure:

        This is the function which will ultimately be registered with the
        multiplexer hence the arguments it expects. Its job is to read a line from
        the client when the multiplexer said the client was readable and then store
        the line into the line-queue. If we have read a line, we immediately
        register the write-a-line handler with the multiplexer since we need to
        know when the client will be ready to accept it. If it turns out there is
        more data stored than the high-water mark we set, we unregister the read
        handler so we don't continue to keep reading data. If we get END-OF-FILE,
        but there is nothing left to write, then this handler performs a small
        optimization and closes the socket to the client and unregisters
        everything. This prevents a needless loop through the multiplexer in this
        case.

        The handling of END-OF-FILE is interesting in that we unregister the read
        handler, since we won't need it anymore, and mark that we've seen the
        END-OF-FILE. At this point, the only thing the multiplexer has to do with
        respect to this client is to write all of the lines stored in the
        line-queue out to the client and close the connection to the client.

        Of the various conditions that can be signaled, the
        SOCKET-CONNECTION-RESET-ERROR condition is the one which will shut down the
        whole connection by removing all handlers in the multiplexer for this
        client and ultimately throw away any in-flight data.

    <example ex7-buffer:ex-1>

2. The write-a-line closure:

        This function is somewhat symmetrical to read-a-line. It will register and
        unregister itself or the read handler based upon how much data is available
        to read/write. If the END-OF-FILE is seen and there is nothing left to
        write, it will close the connection to the client and unregister
        everything.

    <example ex7-buffer:ex-2>

3. The returned closure, which represents the io-buffer:

        This is the actual closure returned by make-ex7-io-buffer and which is used
        to gain access into the read-a-line and write-a-line functions. It takes a
        single argument, either the keywords :read-a-line or :write-a-line, and
        returns a reference to either internal function.

    <example ex7-buffer:ex-3>

While this server still uses blocking I/O, we've laid the foundations for
nonblocking I/O and memory storage enforcement. The foundations specifically
are separating the read/write handlers into different pieces and having shared
lexical bindings between them.

Echo Server IPV4/TCP: ex8-server.lisp
-------------------------------------

This server uses nonblocking I/O and the multiplexer to concurrently talk to
the clients.

Architecturally, it is very similar to ex7-server, but the io-buffer for this
server is implemented with much different internals. Whereas in ex7-server
reading from a client used the stream function READ-LINE, writing used the
stream function FORMAT, and the strings from the client were kept in a queue,
now we use RECEIVE-FROM and SEND-TO along with an array of unsigned-bytes as a
buffer to read/write actual bytes from the socket.

Accessing the socket through the stream API is different than doing it through
the almost raw socket API which we are about to use.  RECEIVE-FROM and SEND-TO
are not part of the stream interface. They are a lower level API in IOLib being
closer to the underlying OS abstraction and as a consequence have a somewhat
different set of conditions that they can signal.  These different conditions
have the form isys:<unix-errno-name> like: isys:epipe, isys:ewouldblock, etc.
There is some intersection with the condition names signaled by the stream API,
such as: SOCKET-CONNECTION-RESET-ERROR, and SOCKET-CONNECTION-REFUSED.

[TODO figure out complete list!]

An example of the ramifications of this API is RECEIVE-FROM. Comparing against
the stream interface whose READ-LINE will signal an END-OF-FILE when the
reading socket has been closed by the client, the function RECEIVE-FROM will
return 0, signifying the end of file. The stream function FORMAT will signal
HANGUP if it tries to write to a socket where the client has gone away. SEND-TO
might not signal, or otherwise produce, any error at all when writing to a
socket where the client has gone away--usually it is on the next RECEIVE-FROM
that it is discovered the client went away. The bytes that SEND-TO wrote simply
vanish!

With IOLib, it may surprise you to be told that all underlying fds in the
previous examples have been nonblocking! This is why we specified :wait t for
ACCEPT-CONNECTION and CONNECT.

The IOLib library internally ensures that the stream interface blocks according
to the requirements of ANSI Common Lisp. However, when we use SEND-TO and
RECEIVE-FROM we automatically gain the benefit of the non-blocking status on
the underlying fd. This is why in this example we don't explicitly set the
underlying fd to non-blocking status--it already is!

The server code itself is described as a difference from ex7-server, but the
io-buffer for this nonblocking server (in file ex8-buffer.lisp) will be
described in its entirety. Also, this server honors the batch input requirement
from example client ex-5b-client, which you should use against this server.

The ex8-server codes:

0. The listener handler (first half):

        Accept and store the client connection.

    <example ex8-server:ex-0>

1. The listener handler (second half):

        Like ex7-server, we register the read and write handlers. Notice though
        that we changed the keywords to the io-buffer closure to be
        :read-some-bytes and :write-some-bytes. This better represents what the
        io-buffer is actually doing.

    <example ex8-server:ex-1>

The rest of the server is extremely similar to ex7-server.

Now, we'll show the io-buffer specific to ex8-server.

0. The internal state of the io-buffer closure:

        The binding echo-buf is an unsigned-byte array of size max-bytes.  This is
        where data from the client is stored before it is written back to the
        client.

        The binding read-index keeps track of the beginning of the empty space in
        the echo-buf buffer where more data could be stored during a read.

    The binding write-index keeps track of how much data has been written to
    the client. It moves towards read-index, and when it has the same value as
    read-index it means that there is no data left to write to the client.

        The bindings read-handler-registered and write-handler-registered allow the
        io-buffer to know when it has registered a handler for reading and writing
        data.

        The binding eof-seen marks when the client has closed its write connection
        to the server. The server will push out all data to the client, then close
        socket to the client.

    <example ex8-buffer:ex-0>

1. Reading bytes form the client:

        In this function, we will convert the return value 0 of RECEIVE-FROM on the
        read of a closed socket into a signaled END-OF-FILE condition to keep the
        structure of our code similar to what has transpired before. Once we read
        some bytes, we increment the read-index pointer and ensure to register a
        write handler to write the data back out. We optimize the writing process a
        little bit and try to write the data out immediately without checking to
        see if the socket is ready. Then if there is no more room in the echo-buf
        array, we unregister ourselves so we don't try and read more data from the
        client until we are ready to accept it (by having written all of the data
        back to the client). We mark the END-OF-FILE flag and unregister the read
        handler if we see the client has closed its connection. We optimize the
        knowledge that if we have no more data to write we just close the
        connection to the client.

    <example ex8-buffer:ex-1>

2. Writing bytes to the client:

        While there are more bytes to write, we write them, keeping track of how
        much we wrote. Once we are out of data to write, we unregister the write
        handler, since we don't want to be called unnecessarily--usually the client
        socket is always ready to write. If we've seen the eof marker and are out
        of data, we close the client connection and are done. If we haven't seen
        it, then we determine if we are at the end of the buffer, if so, we reset
        the indices to the beginning.  Either way, we re-register the read handler
        to acquire more data.

        We handle some new conditions here: isys:ewouldblock is needed because
        sometimes the underlying OS will mark an fd as ready to write when in fact
        it isn't when we get around to writing it. We might also see this condition
        when we tried to optimize the write of the data in the read handler since
        we did it outside of the multiplexer--this is idiomatic and saves a trip
        through the multiplexer more often than not. Seeing isys:ewouldblock simply
        aborts the write and we'll try again later. Under some conditions, send-to
        will signal an isys:epipe error, which means the client closed its
        connection. It is similar to a HANGUP condition in a format call with the
        stream API. We treat it similarly to a HANGUP.

    <example ex8-buffer:ex-2>

3. The returned closure of the io-buffer:

        Much like make-ex7-io-buffer, we return one of the internal closures which
        are appropriate for reading or writing by the multiplexer.

    <example ex8-buffer:ex-3>


Future Directions
-----------------

Of course, more information should go into this tutorial, such as non-blocking
connects/accepts, urgent TCP data, UDP examples, and IPV6. As time permits or
contributions come in, these will be added.

Appendix A
----------

This holds a rough approximation between the sources in this tutorial and the
original sources in the network programming book by Stevens mentioned in the
beginning of the tutorial. Aspects about the implementation of each client or
server are summarized here.

The Clients
-----------

Figure 1.5, page 6
- ex1-client: Blocking I/O, daytime client, C Style

Figure 1.5, page 6
- ex2-client: Blocking I/O, daytime client, Lisp Style

Figure 1.5, page 6
- ex3-client: ex2-client, but with much more error handling

Figure 5.4, 5.5, page 114, 115
- ex4-client: Blocking I/O, line oriented

Figure 6.9, page 157
- ex5a-client: I/O multiplexing with iolib, line oriented, blocking I/O
    - note: since this is still blocking I/O, I'm using *standard-input*
    and friends. Also note, with batch input, it will close the socket with
    in-flight data still present which is incorrect.

Figure 6.13, page 162
- ex5b-client: Same as ex5a-client EXCEPT shutdown is called when the input
    reaches end-of-file as to prevent in flight data from being destroyed
    on the way to the server.

The servers
-----------

Figure 4.11, page 101
- ex1-server: Iterative, blocking I/O daytime server, C Style, no
    error handling, one shot, line oriented

Figure 4.11, page 101
- ex2-server: Iterative, blocking I/O daytime server, Lisp Style,
    no error handling, loop forever, line oriented

Figure 4.11, page 101
- ex3-server: daytime server, ex2-server, but with error handling, line oriented

Figure 4.13, page 105
- ex4-server: daytime server, concurrent, blocking I/O, line oriented

Figure 5.2, 5.3, page 113, 114
- ex5-server: Concurrent, blocking I/O, echo server, line oriented

Figure 6.21,6.22 page 165,166
- ex6-server: I/O multiplexing of clients with iolib, line oriented,
    blocking I/O

Figure 6.21, 6.22 page 165,166
- ex7-server, ex7-buffer: individual I/O handlers for read/write,
    I/O multiplexing of clients with iolib, line oriented, blocking I/O,
    has problem with denial of service, page 167.

Figure 15.3, 15.4, 15.5 page 400-403
- ex8-server, ex8-buffer: nonblocking I/O, event-dispatch, send-to, receive-from